Design and math references for the Vinv monorepo. These explain how the
pipeline reasons — the engines themselves are documented in each package's
own README.md.
| Document | What it covers | Who it's for |
|---|---|---|
| learning.md | The reward, propensity, and gating math that actually ships — every claim carries a file:line reference, plus measured off-policy results on this repo's own telemetry. |
Start here. Anyone who wants to understand how Vinv learns from episodes. |
| optimization-ontology.md | The concept graph behind the optimization loop (detection → board → dispatch → verdict → learning): every node is a real on-disk artifact with its writers, readers, and expiry rule. | Contributors working on the optimization pipeline, or agents walking a repo's optimization state. |
| testing-ontology.md | The concept graph behind behavioral testing: what artifacts must exist, what produces each, and the top-to-bottom walk order — framework-agnostic. | Contributors working on exerciser/identification, or agents reasoning about test coverage. |
learning.md is the entry point; the two ontology documents are walkable maps
of the on-disk artifacts each loop reads and writes.